Subspace Based Identification And Its Application

Most control methods are based on mathematical model. It is nearly impossible to build the mathematical model from mechanism. The demands of modelling process by system identification are focused especially for multi-variables process. The identification based subspace method has been paid more attention in the last two decades. In this dissertation, four issues of subspace identification methods will be studied: closed-loop subspace identification, continuous-time subspace identification, recursive subspace model identification and the application of subspace identification in industrial process.1. Closed-loop subspace identification becomes one of the knotty issues in subspace identification. The closed-loop subspace identification method based on orthogonal decomoposition and principal component analysis is proposed in this paper. A comparison between the proposed methods with other subspace methods based on principal component analysis is made to figure out the biased estimates of other subspace methods based on principal component analysis. And then the proposed algorithm is summed up into the framework of subspace identification based on principal component analysis and instrumental variable. Five different instrumental variables are proposed for this type of closed-loop identification methods. These instrumental variables are compared by numerical simulations. The results show the instrumental variable included the more past input and output could achieve the better performance.2. The continuous-time identification based on subspace method for EIV and closed-loop has not received much attention. The main reason is that the discrete-time identification problems have not yet been solved. In this paper, we solved the continuous-time identification for EIV based on the linear filter method and principal component analysis. Furthermore, the identification method for EIV is extended for continuous-time identification under closed-loop. The numerical simulation and benchmark simulation show the good performance of the proposed algorithms.3. The closed-loop case has not been solved since the lack of appropriate offline closed-loop subspace identification. The recursive subspace identification method based on high order ARX model is proposed. The closed-loop identification problem is solved through high order ARX model. The recursive update of singular value decomposition is transferred to subspace trackning. Numerical simulations show the performance of the algorithm in tracking the time-invariant parameters and time-variant parameters.4. The identification procedure of top gas pressure turbine unit system is proposed. The stability of top gas pressure in blast furnace is the key technology of TRT project. It is difficult to build the model from the first principles directly. In this paper, the continuous-time subspace identification method is applied to set up the state space model of top gas pressure. The simulation results show the identified model fits the true dynamic model well.